A heterogeneous parallel accelerator for image analysis and radar signal processing

The highly specialised Neural Accelerator Board (NAB), and the Shiva (a reconfigurable multiprocessor) are two architectures designed to augment general purpose workstations. The NAB uses custom VLSI components to perform weighted 10-bit fixed-point multiply/accumulate operations at a total rate of 20 giga-operations per second. The Shiva uses Intel i860 microprocessors to achieve a good balance of floating point, graphics, and integer performance. The two can be combined into a heterogeneous, reconfigurable multiprocessor, which has applications in image analysis and radar signal processing. This paper outlines the benefits of heterogeneity and reconfigurability. A description is given of the architectures of the NAB and the Shiva, with details of the proposed applications